Fix unistd.exec_argv0_null for new kernels.
There are other options here (see the code comment for details), but
this is the least effort/least disruptive for now.
Bug: 228898932, 227498625
Test: treehugger
Change-Id: I33be6fbfc022238de2f1846a69af1e712a9d6391
(cherry picked from commit bb1cc5a82c8e19d5f4231988ba48ce3de43ff3ed)
diff --git a/tests/unistd_test.cpp b/tests/unistd_test.cpp
index 7d1e612..19f58e3 100644
--- a/tests/unistd_test.cpp
+++ b/tests/unistd_test.cpp
@@ -1515,11 +1515,21 @@
}
TEST(UNISTD_TEST, exec_argv0_null) {
- // http://b/33276926
+ // http://b/33276926 and http://b/227498625.
+ //
+ // With old kernels, bionic will see the null pointer and use "<unknown>" but
+ // with new (5.18+) kernels, the kernel will already have substituted the
+ // empty string, so we don't make any assertion here about what (if anything)
+ // comes before the first ':'.
+ //
+ // If this ever causes trouble, we could change bionic to replace _either_ the
+ // null pointer or the empty string. We could also use the actual name from
+ // readlink() on /proc/self/exe if we ever had reason to disallow programs
+ // from trying to hide like this.
char* args[] = {nullptr};
char* envs[] = {nullptr};
ASSERT_EXIT(execve("/system/bin/run-as", args, envs), testing::ExitedWithCode(1),
- "<unknown>: usage: run-as");
+ ": usage: run-as");
}
TEST(UNISTD_TEST, fexecve_failure) {
